CN112541912A - Method and device for rapidly detecting saliency target in mine sudden disaster scene - Google Patents

Abstract

The invention relates to a rapid detection method and device for a salient target in a mine sudden disaster scene, belonging to the technical field of computer vision, and solves the problems of poor detection accuracy and slow speed of the existing salient target detection method. The method includes: acquiring an input image about a mine sudden disaster scene; constructing a Unet network, inputting the input image into the Unet network, and obtaining a segmentation map image corresponding to the input image; constructing a fully convolutional FCN network, and inputting the input image into the fully convolutional FCN network , obtain the saliency map image corresponding to the input image; fuse the segmentation map image corresponding to the input image and the saliency map image corresponding to the input image to obtain the fusion map image, and obtain the local map based on the fusion map image; obtain the final image based on the local map salient goals. The rapid detection of salient targets in the mine sudden disaster scene is realized, and the accuracy and speed of salient target detection are improved.

Description

Rapid detection method and device for salient objects in mine sudden disaster scene

technical field

The invention relates to the technical field of computer vision, in particular to a method and device for rapid detection of salient objects in a mine sudden disaster scene.

Background technique

Under the special industrial and mining environment, the video images collected in coal mines have low illuminance, large changes, large background noise and uneven noise distribution. In addition, some underground videos in coal mines include both video images captured by fixed cameras and vehicle-mounted cameras. The influence of the all-weather artificial lighting environment in the coal mine and factors such as dust and humidity make the image acquisition accuracy poor, which greatly affects the presentation accuracy of the scene scene and the remote active early warning linkage effect, and limits the high-precision acquisition and target of mine video. identify applications,

The traditional coal mine video monitoring system can only record the monitoring scene, which requires the staff to observe carefully and continuously, not only requires a high degree of attention, but also cannot quickly alarm and deal with the accident when the accident occurs. The particularity of underground video in coal mines brings great difficulties to the automatic detection of personnel, and also restricts the current methods in the rapid detection and response linkage of salient targets in sudden underground disaster scenarios.

Most of the current saliency target detection methods are detection methods based on deep learning neural networks to obtain deep salient features. The existing problems include: First, target detection mostly focuses on the position of the target center, and there is a certain degree of edge processing and attention. Defects; second, for images with multiple saliency targets, the correlation between each target is not used in saliency reasoning, which makes the existing saliency target detection methods have poor detection accuracy and slow speed.

SUMMARY OF THE INVENTION

In view of the above analysis, the embodiments of the present invention aim to provide a rapid detection method and device for a salient target in a mine sudden disaster scene, so as to solve the problem of poor detection accuracy and slow speed of the existing salient target detection method The problem.

On the one hand, an embodiment of the present invention provides a rapid detection method for a salient target in a mine sudden disaster scene, including the following steps:

Obtain input images of mine sudden disaster scenarios;

Constructing a Unet network, inputting the input image into the Unet network, to obtain a segmentation map image corresponding to the input image, and the segmentation map image includes a plurality of segmentation targets;

Constructing a fully convolutional FCN network, and inputting the input image into a fully convolutional FCN network to obtain a saliency map image corresponding to the input image;

fusing the segmentation map image corresponding to the input image and the saliency map image corresponding to the input image to obtain a fusion map image, and obtain a local map based on the fusion map image;

The final saliency target is obtained based on the local map.

Further, the Unet network includes an encoder and a decoder;

The encoder includes two first networks and four second networks connected in sequence, wherein the first network includes a convolution layer, a batch normalization layer and an activation layer connected in sequence; the second network includes a sequential connection The max pooling layer and the two first networks;

The decoder includes four third networks and one convolutional layer connected in sequence, wherein the third network includes a transposed convolutional layer and two first networks connected in sequence.

Further, the fully convolutional FCN network includes three parts; wherein,

The first part contains five convolutional layers, each of which is followed by a pooling layer;

The second part contains three convolutional layers connected in sequence;

The third part is used to perform an upsampling operation on the output of the second part, and the upsampling operation is implemented based on a pooling layer and a deconvolution layer.

Further, take each segmentation target in the fusion map image as a node, and use the KD nearest neighbor algorithm to calculate the distance and relative position between any two nodes;

Take each node as the central node, obtain the two nodes with the shortest and the second shortest distance from the central node as the one-hop node of the central node, and take the two nodes with the shortest and the second shortest distance from each of the one-hop nodes as the central node the two-hop node;

A node set is obtained based on the one-hop node and two-hop node of the central node, the neighbor nodes of each node in the node set are obtained, and each node in the node set is connected with the neighbor node in the node set , and obtain the local map corresponding to each segmentation target.

Further, obtaining the final saliency target based on the local map includes the following steps:

Based on the graph convolutional reasoning network GCN, combine all the local graphs to obtain a combined graph, and obtain the edge weight of each node in the combined graph and its corresponding one-hop node;

The final saliency target is obtained based on the edge weight of each node in the combined graph and its corresponding one-hop node.

Further, the graph convolutional reasoning network GCN includes four convolutional layers and two fully connected layers connected in sequence.

Further, the final saliency target is obtained based on the edge weight of each node in the combined graph and its corresponding one-hop node, including the following steps:

Determine whether the edge weight of each node and its corresponding one-hop node is greater than the weight threshold, if so, merge the node and its corresponding one-hop node, if not, do not merge the node and its corresponding one-hop node;

Traverse all nodes in the combined graph to get the final saliency target.

On the other hand, an embodiment of the present invention provides a rapid detection device for a salient target in a mine sudden disaster scene, including:

The data acquisition module is used to acquire input images about the mine sudden disaster scene;

The Unet network building module is used to construct the Unet network, and the input image is input into the Unet network to obtain a segmentation map image corresponding to the input image, and the segmentation map image includes a plurality of segmentation targets;

A fully convolutional FCN network building module is used to construct a fully convolutional FCN network, and the input image is input into the fully convolutional FCN network to obtain a saliency map image corresponding to the input image;

a local map obtaining module, configured to fuse the segmentation map image corresponding to the input image and the saliency map image corresponding to the input image to obtain a fusion map image, and obtain a local map based on the fusion map image;

The saliency target obtaining module obtains the final saliency target based on the local map.

Further, the Unet network includes an encoder and a decoder;

The encoder includes two first networks and four second networks connected in sequence, wherein the first network includes a convolution layer, a batch normalization layer and an activation layer connected in sequence; the second network includes a sequential connection The max pooling layer and the two first networks;

The decoder includes four third networks and one convolutional layer connected in sequence, wherein the third network includes a transposed convolutional layer and two first networks connected in sequence.

Further, the fully convolutional FCN network includes three parts; wherein,

The first part contains five convolutional layers, each of which is followed by a pooling layer;

The second part contains three convolutional layers connected in sequence;

The third part is used to perform an upsampling operation on the output of the second part, and the upsampling operation is implemented based on a pooling layer and a deconvolution layer.

Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

1. A rapid detection method for salient objects in mine sudden disaster scenes, which fuses the segmentation map image obtained by Unet network and the saliency map image obtained by fully convolutional FCN network to obtain a fusion map image, and then based on the fusion map The image obtains multiple local maps, and finally processes the multiple local maps through the graph convolution inference network GCN to obtain the final saliency target. The method is simple, easy to implement, considers the relevance of each segmentation target, improves the accuracy of the obtained saliency target, improves the detection speed, and has high practical value.

2. The segmentation map image corresponding to the input image is obtained through the Unet network, which provides technical support and basis for image fusion and salient target detection in the later stage. At the same time, the segmentation map image contains multiple segmentation targets, and the segmentation targets are processed on the basis of the later fusion image to obtain a local map, considering the relationship between the segmentation targets, which is beneficial to improve the accuracy of the obtained saliency targets.

3. Take each segmentation target in the fusion map image as a central node to obtain the local graph corresponding to each central node, and consider the connection between each segmentation target, which is conducive to enhancing the effect of the edge, so that the final obtained The saliency target is more accurate.

In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination solutions. Additional features and advantages of the invention will be set forth in the description which follows, and some of the advantages may become apparent from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by means of particularly pointed out in the description and drawings.

Description of drawings

The drawings are for the purpose of illustrating specific embodiments only and are not to be considered limiting of the invention, and like reference numerals refer to like parts throughout the drawings.

1 is a framework diagram of a method for rapid detection of salient objects in a mine sudden disaster scene in one embodiment;

2 is a flowchart of a method for rapid detection of salient objects in a mine sudden disaster scene in one embodiment;

Fig. 3 is Unet network structure diagram in one embodiment;

Fig. 4 is a fully convolutional FCN network structure diagram in one embodiment;

5 is a structural diagram of a rapid detection device for salient objects in a mine sudden disaster scene in one embodiment;

FIG. 6 is a schematic diagram of a hardware structure of an electronic device for implementing the method for rapid detection of a salient target in a mine sudden disaster scene provided by an embodiment of the present invention in one embodiment.

Reference number:

100-Data acquisition module, 200-Unet network building module, 300-Fully convolutional FCN network building module, 400-Local graph acquisition module, 500-Saliency target acquisition module, 610-Processor, 620-Memory, 630-Input Device, 640 - Output Device.

Detailed ways

The preferred embodiments of the present invention are specifically described below with reference to the accompanying drawings, wherein the accompanying drawings constitute a part of the present application, and together with the embodiments of the present invention, are used to explain the principles of the present invention, but are not used to limit the scope of the present invention.

Most of the current saliency target detection methods are detection methods based on deep learning neural networks to obtain deep salient features. The existing problems include: First, target detection mostly focuses on the position of the target center, and there is a certain degree of edge processing and attention. Defects; second, for images with multiple saliency targets, the correlation between each target is not used in saliency reasoning. To this end, the present application proposes a method and device for rapid detection of salient objects in a mine sudden disaster scene. As shown in Figure 1, the segmentation map image corresponding to the input image is obtained through the Unet network, and obtained through the fully convolutional FCN network. Input the saliency map image corresponding to the image, and fuse the segmentation map image and the saliency map image to obtain a fusion map image, then obtain multiple local maps based on the fusion map image, and finally pass the graph convolution inference network GCN to the multiple local maps. After processing, the connection between each target is obtained, and the final saliency target is obtained according to the connection between the targets. The method is simple, easy to implement, considers the relevance of each segmentation target, improves the accuracy of the obtained saliency target, improves the detection speed, and has high practical value.

A specific embodiment of the present invention discloses a rapid detection method for a salient target in a mine sudden disaster scene, as shown in FIG. 2 , including the following steps S1 to S5.

Step S1, acquiring an input image about a mine sudden disaster scene. Specifically, the input image of the mine sudden disaster scene can be obtained from the video image captured by the fixed camera in the coal mine, and can also be obtained from the video image captured by the vehicle-mounted camera in the coal mine.

Step S2, constructing a Unet network, inputting the input image into the Unet network, and obtaining a segmentation map image corresponding to the input image, and the segmentation map image includes multiple segmentation targets. Specifically, as shown in Figure 3, the Unet network includes an encoder and a decoder; where,

The encoder includes two first networks and four second networks connected in sequence, wherein the first network includes a convolutional layer Conv, a batch normalization layer BN and an activation layer Relu connected in sequence; the second network includes a max pool connected in sequence Layer Maxpool and two first networks. In detail, the input image generates 64-dimensional feature maps through two first networks, and the 64-dimensional feature maps generate 1024-dimensional feature maps through four second networks.

The decoder includes four third networks connected in sequence and one convolutional layer, wherein the third network includes a transposed convolutional layer UpConv connected in sequence and two first networks. In detail, the 1024-dimensional feature map is sampled into a 64-dimensional feature map of the original image size through four third networks, and the intermediate feature map generated by the encoder is spliced to the feature map generated by the decoder using skip connections; A convolutional layer generates segmented images from the concatenated 64-dimensional feature maps.

The segmentation map image corresponding to the input image is obtained through the Unet network, which provides technical support and basis for image fusion and saliency target detection in the later stage. At the same time, the segmentation map image contains multiple segmentation targets, and the segmentation targets are processed on the basis of the later fusion image to obtain a local map. Considering the relationship between each segmentation target, it is beneficial to improve the accuracy of the obtained saliency target.

Step S3, constructing a fully convolutional FCN network, and inputting the input image into the fully convolutional FCN network to obtain a saliency map image corresponding to the input image. The fully convolutional FCN network consists of three parts; as shown in Figure 4, the first part contains five convolutional layers, each convolutional layer is connected with a pooling layer; among them, the pooling layer connected after each convolutional layer Used to change the size of the image to 1/2 the size of the previous layer.

The second part consists of three convolutional layers connected in sequence, changing the dimensions of the image to 4096, 4096, and 1000 dimensions, respectively.

The third part is used to perform an up-sampling operation on the output of the second part. The up-sampling operation is implemented based on the pooling layer and the deconvolution layer, which is embodied in the up-pooling and deconvolution operations.

The saliency map image corresponding to the input image is obtained through the fully convolutional FCN network, which provides technical support and basis for the later image fusion and saliency target detection. Through the cooperation of the Unet network and the fully convolutional FCN network, it is beneficial to improve the accuracy and detection speed of the obtained salient objects.

Step S4, fuse the segmentation map image corresponding to the input image and the saliency map image corresponding to the input image to obtain a fusion map image, and obtain a local map based on the fusion map image. Specifically, by fusing the segmentation map image with the saliency map image, a fused map image can be obtained, and by processing the fused map image, multiple local maps can be obtained.

Preferably, obtaining a local map based on the fusion map image includes the following steps:

Step S401 , take each segmentation target in the fusion map image as a node, and use the KD nearest neighbor algorithm to calculate the distance and relative position between any two nodes. Specifically, inputting the input image into the Unet network can obtain a segmentation map image. The segmentation map image contains multiple segmentation targets. Therefore, the fusion map image also contains multiple segmentation targets. Each segmentation target is regarded as a node. The neighbor algorithm calculates the distance and relative position between any two nodes. In detail, the principle of the KD nearest neighbor algorithm is: first find the leaf node containing the target point; then start from the leaf node, return to the parent node at a time, and continuously find the node closest to the target point, when it is determined that there is no closer node. When the node stops, the distance and relative position between any two nodes can be obtained.

Step S402, take each node as the central node, obtain two nodes with the shortest and second shortest distances from the central node as the one-hop nodes of the central node, and take the shortest and second shortest two nodes from each one-hop node as the center The node's two-hop node.

In step S403, a node set N is obtained based on the one-hop node and the two-hop node of the central node, and the three neighbor nodes Ne of each node in the node set are obtained, and each node in the node set is compared with its neighbors in the node set. The nodes are connected to obtain the local graph corresponding to each segmentation target. Specifically, after obtaining the three neighbor nodes Ne of each node in the node set, it is judged whether each neighbor node Ne is in the node set N, if the neighbor node Ne is in the node set N, the node and its in the node set The adjacent nodes are connected to obtain the local graph corresponding to the segmentation target. By traversing each central node, the local graph corresponding to each segmentation target can be obtained.

Taking each segmentation target in the fusion map image as a central node to obtain the local graph corresponding to each segmentation target, considering the connection between each segmentation target, it is beneficial to enhance the effect of the edge and make the final obtained saliency Target accuracy is higher.

Step S5, obtaining the final saliency target based on the local graph, including the following steps:

Step S501 , combine all local graphs based on the graph convolutional reasoning network GCN to obtain a combined graph, and obtain the edge weight of each node in the combined graph and its corresponding one-hop node. Specifically, the graph convolutional reasoning network GCN includes four convolutional layers and two fully connected layers connected in sequence, and the activation function used in the convolutional layer in the GCN network is PReLU. In detail, the graph convolutional reasoning network GCN obtains the edge weight calculation formula of each node in the combined graph and its corresponding one-hop node as follows:

Y=σ[(X||GX)W]

In the above formula, Y is the edge weight of each node in the combined graph and its corresponding one-hop node, X is the feature matrix of the local graph, G is the aggregation matrix, W is the weight matrix of the graph convolution inference network GCN, σ() represents a nonlinear activation function, where G=Λ ^-1/2 AΛ ^-1/2 , Λ represents a diagonal matrix, and A represents an adjacency matrix.

Step S502, obtaining the final saliency target based on the edge weight of each node in the combined graph and its corresponding one-hop node, including the following steps:

Determine whether the edge weight of each node and its corresponding one-hop node is greater than the weight threshold, if so, merge the node and its corresponding one-hop node, if not, do not merge the node and its corresponding one-hop node; traverse the combined graph All nodes of , get the final saliency target.

Specifically, based on the above graph convolutional reasoning network GCN, all local graphs can be combined to obtain a combined graph, and the edge weight of each node in the combined graph and its corresponding one-hop node can be obtained. Next, it is judged whether the edge weight of each node and its corresponding one-hop node is greater than the weight threshold, wherein the weight threshold is manually set in the actual situation. The final saliency target can be obtained by merging the edge weights of each node in the combined graph and its corresponding one-hop node, and traversing all the nodes in the combined graph.

Compared with the prior art, the method and device for rapid detection of salient targets in a mine sudden disaster scene provided by this embodiment are obtained by fusing the segmentation map image obtained by the Unet network and the saliency map image obtained by the fully convolutional FCN network. Fuse the map image, and then obtain multiple local maps based on the fusion map image, and finally process the multiple local maps through the graph convolution inference network GCN to obtain the connection between each target. sexual goals. The method is simple, easy to implement, considers the relevance of each segmentation target, improves the accuracy of the obtained saliency target, improves the detection speed, and has high practical value.

Another specific embodiment of the present invention discloses a rapid detection device for salient objects in a mine sudden disaster scene, as shown in FIG. 5 , including:

The data acquisition module 100 is used for acquiring the input image about the mine sudden disaster scene;

The Unet network construction module 200 is used to construct the Unet network, and the input image is input into the Unet network to obtain a segmentation map image corresponding to the input image, and the segmentation map image includes a plurality of segmentation targets;

The fully convolutional FCN network construction module 300 is used to construct a fully convolutional FCN network, input the input image into the fully convolutional FCN network, and obtain a saliency map image corresponding to the input image;

The local map obtaining module 400 is used to fuse the segmentation map image corresponding to the input image and the saliency map image corresponding to the input image to obtain a fusion map image, and obtain a local map based on the fusion map image;

The saliency target obtaining module 500 obtains the final saliency target based on the local map.

Since the realization principle of the rapid detection device for salient objects in the mine sudden disaster scene is the same as the realization principle of the aforementioned rapid detection method for the salient objects in the mine sudden disaster scene, it will not be repeated here.

Referring to FIG. 6 , another embodiment of the present invention further provides an electronic device for implementing the method for rapid detection of a salient target in a mine sudden disaster scene in the above embodiment. The electronic equipment includes:

One or more processors 610 and a memory 620, one processor 610 is taken as an example in FIG. 6 .

The electronic device of the method for rapid detection of a salient target in a mine sudden disaster scene may further include: an input device 630 and an output device 640 .

The processor 610, the memory 620, the input device 630, and the output device 640 may be connected by a bus or in other manners, and the connection by a bus is taken as an example in FIG. 6 .

As a non-volatile computer-readable storage medium, the memory 620 can be used to store non-volatile software programs, non-volatile computer-executable programs and modules, such as in the mine sudden disaster scenario in the embodiment of the present invention. Program instructions/modules (units) corresponding to the method for rapid detection of salient objects. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, to implement the icon display method in the above method embodiment.

The memory 620 may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store the quantity information of the acquired reminder items of the application program, and the like. Additionally, memory 620 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory located remotely from the processor 610, which remote memory may be connected via a network to the processing device operated by the listing. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The input device 630 may receive the input numerical or character information, and generate key signal input related to the user setting and function control of the rapid detection device of the salient object in the mine sudden disaster scene. The output device 640 may include a display device such as a display screen.

The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, execute the method for rapid detection of a salient target in a mine sudden disaster scenario in any of the above method embodiments .

The above product can execute the method provided by the embodiments of the present invention, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference may be made to the methods provided by the embodiments of the present invention.

Electronic devices of embodiments of the present invention may exist in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions, and its main goal is to provide voice and data communication. Such terminals include: smart phones (eg iPhone), multimedia phones, functional phones, and low-end phones.

(2) Ultra-mobile personal computer equipment: This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has the characteristics of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, such as iPads.

(3) Portable entertainment equipment: This type of equipment can display and play multimedia content. Such devices include: audio and video players (eg iPod), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

(4) Server: A device that provides computing services. The composition of the server includes a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but due to the need to provide highly reliable services, the processing capacity, stability , reliability, security, scalability, manageability and other aspects of high requirements.

(5) Other electronic devices with reminder record function.

The device embodiments described above are only illustrative, wherein the units (modules) described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, That is, it can be located in one place, or it can be distributed to multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

An embodiment of the present invention provides a non-transitory computer-readable storage medium, where computer-executable instructions are stored in the computer storage medium, wherein when the computer-executable instructions are executed by an electronic device, Execute the method for rapid detection of salient objects in a mine sudden disaster scene in any of the above method embodiments.

An embodiment of the present invention provides a computer program product, wherein the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are When executed by an electronic device, the electronic device is made to execute the method for rapid detection of a salient target in a mine sudden disaster scene in any of the above method embodiments.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic A disc, an optical disc, etc., includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or some parts of the embodiments.

The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited to this. Substitutions should be covered within the protection scope of the present invention.

Claims (10)

1. A method for rapidly detecting a significant target in a mine sudden disaster scene is characterized by comprising the following steps:

acquiring an input image about a mine sudden disaster scene;

constructing a Unet network, inputting the input image into the Unet network to obtain a segmentation mapping image corresponding to the input image, wherein the segmentation mapping image comprises a plurality of segmentation targets;

constructing a full convolution FCN network, inputting the input image into the full convolution FCN network, and obtaining a significance mapping image corresponding to the input image;

fusing the segmentation mapping image corresponding to the input image with the saliency mapping image corresponding to the input image to obtain a fused mapping image, and obtaining a local image based on the fused mapping image;

and obtaining a final significance target based on the local graph.

2. The method of claim 1, wherein the Unet network comprises an encoder and a decoder;

the encoder comprises two first networks and four second networks which are connected in sequence, wherein the first networks comprise a convolution layer, a batch normalization layer and an activation layer which are connected in sequence; the second network comprises a maximum pooling layer and two first networks which are connected in sequence;

the decoder comprises four third networks and a convolutional layer which are sequentially connected, wherein the third networks comprise a transposed convolutional layer and two first networks which are sequentially connected.

3. The method for rapidly detecting a salient object in a mine sudden-disaster scenario according to claim 2, wherein the fully-convoluted FCN network comprises three parts; wherein,

the first part comprises five convolution layers, and a pooling layer is connected behind each convolution layer;

the second part comprises three convolution layers which are connected in sequence;

and the third part is used for performing an up-sampling operation on the output of the second part, wherein the up-sampling operation is realized based on the pooling layer and the deconvolution layer.

4. The method for rapidly detecting the saliency target in the mine sudden disaster scene according to claim 1 is characterized in that a local map is obtained based on the fusion mapping image, and the method comprises the following steps:

taking each segmentation target in the fusion mapping image as a node, and calculating by adopting a KD nearest neighbor algorithm to obtain the distance and the relative position between any two nodes;

taking each node as a central node, obtaining two nodes which are shortest and second shortest from the central node as a first-hop node of the central node, and taking the two nodes which are shortest and second shortest from each first-hop node as a second-hop node of the central node;

and obtaining a node set based on the first hop node and the second hop node of the central node, obtaining a neighbor node of each node in the node set, and connecting each node in the node set with the neighbor node in the node set to obtain a local graph corresponding to each segmentation target.

5. The method for rapidly detecting the saliency target in the mine sudden disaster scene according to claim 4 is characterized in that the final saliency target is obtained based on the local map, and the method comprises the following steps:

combining all local graphs based on a graph convolution reasoning network GCN to obtain a combined graph, and acquiring the edge weight of each node in the combined graph and a corresponding one-hop node;

and obtaining a final significance target based on the edge weight of each node in the combined graph and the corresponding one-hop node.

6. The method for rapidly detecting the saliency target in the mine sudden disaster scene as claimed in claim 5, wherein said graph convolution inference network GCN includes four convolution layers and two fully connected layers connected in sequence.

7. The method for rapidly detecting the saliency target in the mine sudden disaster scene according to claim 6, wherein the final saliency target is obtained based on the edge weight of each node in the combined graph and the corresponding one-hop node, and the method comprises the following steps:

judging whether the edge weight of each node and the corresponding one-hop node is greater than a weight threshold value, if so, combining the node and the corresponding one-hop node, and if not, not combining the node and the corresponding one-hop node;

and traversing all nodes in the combined graph to obtain a final saliency target.

8. A device for rapidly detecting a significant target in a mine sudden disaster scene is characterized by comprising:

the data acquisition module is used for acquiring an input image related to a mine sudden disaster scene;

the Unet network construction module is used for constructing the Unet network, inputting the input image into the Unet network, and obtaining a segmentation mapping image corresponding to the input image, wherein the segmentation mapping image comprises a plurality of segmentation targets;

the full-convolution FCN network construction module is used for constructing a full-convolution FCN network, inputting the input image into the full-convolution FCN network and obtaining a significance mapping image corresponding to the input image;

a local map obtaining module, configured to fuse the segmentation mapping image corresponding to the input image and the saliency mapping image corresponding to the input image to obtain a fused mapping image, and obtain a local map based on the fused mapping image;

and the saliency target obtaining module is used for obtaining a final saliency target based on the local map.

9. The apparatus for rapidly detecting a salient object in a mine sudden disaster scene according to claim 8, wherein said Unet network comprises an encoder and a decoder;

the encoder comprises two first networks and four second networks which are connected in sequence, wherein the first networks comprise a convolution layer, a batch normalization layer and an activation layer which are connected in sequence; the second network comprises a maximum pooling layer and two first networks which are connected in sequence;

the decoder comprises four third networks and a convolutional layer which are sequentially connected, wherein the third networks comprise a transposed convolutional layer and two first networks which are sequentially connected.

10. The apparatus for rapidly detecting a salient object in a mine sudden-disaster scenario according to claim 9, wherein the fully-convoluted FCN network comprises three parts; wherein,

the first part comprises five convolution layers, and a pooling layer is connected behind each convolution layer;

the second part comprises three convolution layers which are connected in sequence;

and the third part is used for performing an up-sampling operation on the output of the second part, wherein the up-sampling operation is realized based on the pooling layer and the deconvolution layer.

Images